Tier to Layer correspondences

When parsing an ELAN file, the general assumption is that each TextGrid tier corresponds to an annotation layer, and before being fully processed, correspondences between tiers and layers need to be specified. The formatter tries to select sensible defaults for these correspondences; as a general rule, if the name of the tier matches the name of an existing annotation layer, then the tier will be mapped to the layer with the same name.

If no automatic correspondence is obvious, the formatter assumes that the tier contains the transcript of the speech of one speaker; the tier is mapped to the “utterance” layer, and the tier's Participant attribute (or the tier name if the Participant attribute is blank) is used as the speaker's name/ID.

While determining default tier-to-layer mappings, the following special cases also apply:

• tiers named lines or utterances are mapped to the utterance layer.
• tiers named speaker[s], turn[s], or “utterances” are mapped to the turn (speaker turn) layer.
• tiers with names that include word are mapped to the word layer.

Meta-data

The Content Language attribute of the tiers, if set, is used for setting the transcript language.

The AUTHOR attribute of the transcript, if set, can be mapped to an author/transcriber attribute layer.

The DATE attribute of the transcript, if set, can be mapped to an date attribute layer.

If PROPERTY tags in the .eaf file's XML code include have a NAME attribute that is prefixed metadata:, then the formatter will attempt to parse the meta-data values into corresponding attribute layers. For example:

• <PROPERTY NAME="metadata:location">Flores</PROPERTY> will map to the location annotation layer by default, and create an annotation labelled “Flores”.
• <PROPERTY NAME="metadata:Gender:Anne">F</PROPERTY> will map to the Gender participant annotation layer by default, and tag the participant labelled “Anne” with an annotation labelled “F”.

Conventions for non-speech annotations within the transcript

ELAN has no direct mechanism for marking non-speech annotations in their position within the transcript text. However, LaBB-CAT supports the use of textual conventions in various ways to make certain annotations:

• To tag a word with its pronunciation, enter the pronunciation in square brackets (with no spaces), directly following the word (i.e. with no intervening space), e.g.:\ …this was at Wingatui[wIN@tui]…
• To tag a word with its full orthography (if the transcript doesn't include it), enter the orthography (with no spaces) in round parentheses, directly following the word (i.e. with no intervening space), e.g.:\ …I can't remem~(remember)…
• To insert a noise annotation within the text, enclose it in square brackets (surrounded by spaces so it's not taken as a pronunciation annotation), e.g.:\ …sometimes me [laughs] not always but sometimes…
• To insert a comment annotation within the text, enclose it in curly braces (surrounded by spaces), e.g.:\ …beautifully warm {softly} but its…
• To tag a word as being in a different language, enter the code CS: (for ‘code switch’) followed by the the ISO 639 3-letter code for the language, in square brackets (with no spaces), directly following the word (i.e. with no intervening space), e.g.:\ …me mudé de New[CS:eng] Zealand[CS:eng] en 2004…
• For longer phrases, the code-switch tag can be placed immediately before the first word and immediately after the last word, to mark those and all intervening words as being in a different language. e.g.:\ …has a certain [CS:fre]je ne sais quoi[CS:fre] I think…

The ‘code switch’ example above is a specific case of a general coding mechanism that can be used; any ‘SALT style’ code can be used to tag a word or phrase in this manner, as long as the code:

• is up to three uppercase ASCII letters long,
• is followed by a colon followed by the annotation label with no spaces (if the label is omitted, the layer ID is used for the annotation label),
• is enclosed in square brackets, and
• has no white-space between it and the token it is tagging.

If there are such codes in the transcript, then the code must be assigned to an annotation layer before the transcript is processed; by default a code will be mapped to an annotation layer with the same name (either upper- or lower-case).

For example, given a word-tag layer called ep, the following transcript would create an ep-layer tag labelled “they”:
And them[EP:they] found the frog.

During processing, any of these annotations will be extracted from the transcript text and inserted into corresponding LaBB-CAT layers.

Configuration

The following parameters can be specified for the formatter:

ignoreBlankAnnotations

If true, annotations with no label are skipped.

useConventions

true to use transcript conventions to identify comment, noise, lexical and pronuncation annotations. (see Transcription Conventions).

commentLayer

ID of layer for commentary (see Transcription Conventions),

noiseLayer

ID of layer for background/non-verbal noises (see Transcription Conventions).

lexicalLayer

ID of layer for lexical tags which identify to lexical item if the token orthography doesn't do so (see Transcription Conventions).

pronounceLayer

ID of layer for manual pronunciation tags (see Transcription Conventions).

authorLayer

ID of the transcript attribute layer for the name of transcriber (see Meta-data).

dateLayer

ID of the transcript attribute layer for the document date (see Meta-data).

languageLayer

ID of the transcript attribute layer for the transcript language (see Meta-data).

phraseLanguageLayer

ID of the aligned phrase layer for tagging groups of words as being in a different language (see Transcript Conventions).

minimumTurnPauseLength

Minimum amount of time (in seconds) between two turns by the same speaker, with no intervening speaker, for which the inter-turn pausecounts as a turn change boundary. If the pause is shorter than this, the turns are merged into one.